Skip to main content

Table 5 Results of classification modelling for target variable lymph node status. Correlation-based feature selection (FS) refers to the method described in the Materials and Methods section, incorporating both ICC and Spearman rank correlations assessed in order of feature groups. Full feature selection starts with the features retained by the correlation-based approach and then applies R’s rfe algorithm under cross-validation. Results represent the mean AUC for 5 repetitions of 10-fold cross-validation, with standard deviations in the range 0.14–0.21 and standard error in the mean 0.02–0.03. However, the Individual data AUC values are not normally-distributed, independent random variables, and so these values should be regarded as indicative only and we do not quote an estimated confidence interval

From: “Real-world” radiomics from multi-vendor MRI: an original retrospective study on the prediction of nodal status and disease survival in breast cancer, as an exemplar to promote discussion of the wider issues

Model type

Variables included

AUC (correlation-based FS)

AUC (correlation-based FS + recursive elimination)

SVM

Clinical

0.68

0.71

Random forest

Clinical

0.72

0.74

XGBoost

Clinical

0.68

0.72

Naïve Bayes

Clinical

0.71

0.72

SVM

Radiomics

0.55

0.62

Random forest

Radiomics

0.57

0.64

XGBoost

Radiomics

0.48

0.60

Naïve Bayes

Radiomics

0.65

0.67

SVM

Clinical + Radiomics

0.66

0.70

Random forest

Clinical + Radiomics

0.62

0.74

XGBoost

Clinical + Radiomics

0.56

0.71

Naïve Bayes

Clinical + Radiomics

0.67

0.76